Authors:
Aloïs Rautureau, Tristan Cazenave, Eric Piette

Venue:
Computer and Games (CG), 2026 (accepted)

Topics:
Monte Carlo Tree Search, GRAVE, memory-constrained AI, general game playing, human-like AI

Links: PDF · arXiv

This paper studies how to preserve the strength of GRAVE, a strong Monte Carlo Tree Search variant for General Game Playing, under strict memory constraints.

It introduces three new algorithms: GRAVE2, based on two-level search, GRAVER, based on node recycling, and GRAVER2, which combines both techniques. The goal is to drastically reduce the number of stored nodes while maintaining playing strength.

Experimental results on 9×9 Go show that these methods can match the performance of standard GRAVE while using only a small fraction of the original memory budget, making them promising for deployment in memory-limited environments.

Monte Carlo Tree Search methods are usually designed under the assumption that memory is abundant, which limits their applicability on devices such as smartphones or embedded systems.

This work addresses that limitation by proposing memory-bounded variants of GRAVE that remain competitive despite storing far fewer search nodes. The paper also connects this question to cognitive modelling, since bounded-memory tree search may better reflect how humans reason and plan under limited informational resources.

More broadly, the paper contributes to the development of efficient and cognitively plausible search algorithms for general game playing, linking practical AI constraints with human-like decision-making models.

Rautureau, A., Cazenave, T., Piette, E. (2026). Generalized Rapid Action Value Estimation in Memory-Constrained Environments. Computer and Games (CG). Accepted.

@inproceedings{rautureau2026grave_memory,
  author    = {Rautureau, Alo{\"i}s and Cazenave, Tristan and Piette, Eric},
  title     = {Generalized Rapid Action Value Estimation in Memory-Constrained Environments},
  booktitle = {Computer and Games (CG)},
  year      = {2026},
  note      = {Accepted},
  url       = {https://arxiv.org/abs/2602.23318}
}